{
  "abstract": "Background During exploration missions, astronauts will be away from Earth for extended periods of time. As the distance from Earth increases, real-time communication with flight surgeons and expert medical consultants becomes possible. Accordingly, management of medical problems must be crew-based rather than ground-based, and medical expertise must reside on board. This paradigm shift has been labelled Earth Independent Medical Operations (EIMO). The recent emergence of Chat GPT and other LLM’s has intrigued scientists with the possibility of using these tools for spaceflight Clinical Decision Support, but the reliability of Generative Pretrained Transformer technologies remains questionable due to errors in information recall/retrieval (deemed hallucinations) and hidden biases in training data sets. The system designed and implemented for NASA utilizes three distinct AI agents with different technologies and modes of operation to work in concert to achieve accurate, reliable in situ diagnostic and procedural guidance for expert and novice first responders, with zero remote dependencies.The need for human-centric portable AI, that supports holistic medical autonomy during remote, communications constrained/denied missions is a significant priority for NASA’s future missions, but also for US and NATO military forces in today’s battlefield. The US Army Futures Command, US Navy/USMC and USAF have each selected the platform developed by Tietronix over the past 9 years (codenamed IMCA, or the Intelligent Medical Crew Agent) for a combined AI medical support package to be evaluated in actual large scale simulated combat in 2025 (figure 1A). Furthermore, IMCA was just selected for formal clinical trial funding by the US Congressionally Directed Medical Research Program. By these selections, our team will have an extraordinary opportunity to test, refine and mature these AI agents into a profoundly robust, lightweight package of medical AI tools today that will benefit NASA’s Mars, and other deep space human crewed missions and allied military combat forces (figure 1B).Innovation Over the past 9 years our team has developed and tested multiple generations of AI technology to achieve the goal of a ‘Doc-in-a-box’ spaceflight-compatible tool to support EIMO. To overcome gaps in reliability, our team has developed an integrated system that employs multiple AI agents, each with distinct architecture and roles, to support to achieve immediate and reliable medical support to remote crews with small, portable computers. The current system utilizes three distinct agents, each with differing rolls and levels of ‘guardrail’ that work together supporting crewmember medical autonomy. These are IMCA, a dialogue-based NLP agent to simplify AI Agent interfacing, VULCAN, a spatial computing agent that ‘shows’ operators in 3D how to perform medical procedures and provides instant feedback to ensure proper execution of procedures, and S-Med GPT, a Small Language Model GPT (termed Space Medicine GPT) to support expansive medical knowledge and diagnostic support. Safety and reliability are tantamount and achieved through a process we have designated as ‘Mutual Agent Consensus’ where the Agents work in concert to insure correct diagnoses and correct procedure execution. Finally, these agents have now been dynamically integrated into a single briefcase-sized platform that allows for rapid medical scenario methods/procedures development, testing, validation, and deployment. These integrated functions have been refined with over two hundred individual human factors test sessions, including simulated Mars spacecraft scenarios and combat simulations, resulting in an effective just-in-time medical support system.Abstract A19 Figure 1AIMCA being used in forward casualty careAbstract A19 Figure 1BIMCA V3.0 (NASA version) versus V3.5 (Military Spec version) L to RAcknowledgments The authors would like to thank NASA for their past and ongoing research support of these technologies. This work was supported by NASA SBIR funding under grant numbers NNX16CC522P, NNX17CC12C, 80NSSC120C0541, 80NSSC21C0578, 80NSSC23PB612.Conflict of Interest Statement The authors declare no competing interests. None of the authors have any financial or personal relationships with organizations that could inappropriately influence (bias) their work. All funding sources and affiliations related to this research have been transparently disclosed.",
  "authors": [
    {
      "affiliations": [
        "Tietronix Software Inc., Houston, TX USA 77058"
      ],
      "name": "William R Buras"
    },
    {
      "affiliations": [
        "Tietronix Software Inc., Houston, TX USA 77058"
      ],
      "name": "D Hashem"
    },
    {
      "affiliations": [
        "Tietronix Software Inc., Houston, TX USA 77058"
      ],
      "name": "K Nguyen"
    }
  ],
  "title": "A19 Developing reliable artificially intelligent (AI) agents for autonomous medical operations",
  "uid": "7c6f40ea-6d93-5edb-9d7d-074bcdc2010f"
}